BUG: Load data from a CategoricalIndex for dtype comparison, closes #…

…16627

doc/source/whatsnew/v0.21.0.txt

@@ -100,6 +100,7 @@ Indexing
 
 - When called with a null slice (e.g. ``df.iloc[:]``), the ``.iloc`` and ``.loc`` indexers return a shallow copy of the original object. Previously they returned the original object. (:issue:`13873`).
 - When called on an unsorted ``MultiIndex``, the ``loc`` indexer now will raise ``UnsortedIndexError`` only if proper slicing is used on non-sorted levels (:issue:`16734`).
+- Fixed a bug that prevented joining on a categorical MultiIndex (:issue:`13873`).
 
 
 I/O

pandas/core/reshape/merge.py

@@ -11,7 +11,7 @@
 import pandas.compat as compat
 
 from pandas import (Categorical, Series, DataFrame,
-                    Index, MultiIndex, Timedelta)
+                    Index, MultiIndex, Timedelta, CategoricalIndex)
 from pandas.core.frame import _merge_doc
 from pandas.core.dtypes.common import (
     is_datetime64tz_dtype,
@@ -1441,9 +1441,13 @@ def _factorize_keys(lk, rk, sort=True):
         rk = rk.values
 
     # if we exactly match in categories, allow us to use codes
+    if isinstance(lk, CategoricalIndex):
+        ldata = lk._data
+    else:
+        ldata = lk
     if (is_categorical_dtype(lk) and
             is_categorical_dtype(rk) and
-            lk.is_dtype_equal(rk)):
+            ldata.is_dtype_equal(rk)):
         return lk.codes, rk.codes, len(lk.categories)
 
     if is_int_or_datetime_dtype(lk) and is_int_or_datetime_dtype(rk):

pandas/tests/test_join.py

@@ -192,3 +192,24 @@ def test_inner_join_indexer2():
 
     exp_ridx = np.array([0, 1, 2, 3], dtype=np.int64)
     assert_almost_equal(ridx, exp_ridx)
+
+
+def test_merge_join_categorical_multiindex():
+    # From issue 16627
+    import pandas as pd
+    a = {'Cat1': pd.Categorical(['a', 'b', 'a', 'c', 'a', 'b'],
+                                ['a', 'b', 'c']),
+         'Int1': [0, 1, 0, 1, 0, 0]}
+    a = pd.DataFrame(a)
+
+    b = {'Cat': pd.Categorical(['a', 'b', 'c', 'a', 'b', 'c'],
+                               ['a', 'b', 'c']),
+         'Int': [0, 0, 0, 1, 1, 1],
+         'Factor': [1.1, 1.2, 1.3, 1.4, 1.5, 1.6]}
+    b = pd.DataFrame(b).set_index(['Cat', 'Int'])['Factor']
+
+    c = pd.merge(a, b.reset_index(), left_on=['Cat1', 'Int1'],
+                 right_on=['Cat', 'Int'], how='left')
+    d = a.join(b, on=['Cat1', 'Int1'])
+    c = c.drop(['Cat', 'Int'], axis=1)
+    assert_almost_equal(c, d)